In the framework of the FTS/jMax system for real-time signal processing, a module to estimate spectral envelopes, manipulate them, and apply them to synthesis would widen greatly the range of applications of the system. Flexible, musically controllable manipulation paradigms in real-time using spectral envelopes could be developed using the graphical programming environment of jMax. However, the demands in processing power for real-time application of the spectral envelope handling algorithms--especially of estimation--would have to be evaluated.
The already existing real-time implementation of additive synthesis would benefit greatly from the addition of a noise model for the residual part of a sound, allowing sharp transients like e.g. in the attack of a sound to be reproduced or generated exactly. Also, for many sounds, a noise component is characteristic and always present, as in the turbulence noises of wind instruments.
Another possible application is in the ESCHER system, based on jMax, designed to provide an intuitive and expressive control of sound synthesis in real-time by all types of gestural input devices [WSR98].
The real-time additive synthesis component in ESCHER is based on an n-dimensional timbral parameter space9.1, spanned up by points of different combinations of the n parameters. At each point, a note of a real instrument played with these timbral parameters is subjected to additive analysis, and the resulting partial sets are stored. Now, when playing the ESCHER system, the partial sets are interpolated according to the current timbral parameters.
There are now two possible applications of spectral envelopes in ESCHER: First, data compression could be used, since the amount of additive analysis data is not negligible (remember that a whole note comprising several hundreds to thousands of time-frames of data is stored at each point of the multi-dimensional parameter grid). Storing the development of the partial amplitudes as a spectral envelope, while the partial frequencies are completely determined by the pitch, would drastically reduce the amount of data to be stored.
Second, along the same lines as in section 8.1, the representation of the instrument as spectral envelopes would allow for better manipulation of the sound (so far, only interpolation is possible). These manipulations could include various partial transformations, which will even amount to extrapolating the sound, i.e. creating sound characteristics which are not restricted to that of the instrument that was originally modeled, resulting in more flexibility and expressiveness.